Revealed Preference Tests of Network Formation Models

This paper proposes a revealed preference test of network formation models. Specifically, I consider network formation models where agents are (1) strategic, (2) externalities are confined to within an agent's k-neighborhood, where k can be varied. I show that this model can be tested using observation of a single network. I then derive necessary and sufficient condition under which the observed network is consistent with our strategic models of network formation. This non-parametric test takes the form of an algorithm involving the computation of color-preserving automorphisms of graphs. Building on the theoretical result, the test is implemented to calculate its' statistical power and to the Banerjee et al. (2012)'s social network data